Investigation of Bearing Condition by Means of Robust Linear Regression and Informative Predictors

This study addresses the condition monitoring of rolling bearings by applying robust linear regression to statistically derived features from vibration data. Four datasets of acceleration signals were collected under varying operating conditions: aligned and misaligned bearings at rotational speeds of 1000 rpm and 1500 rpm. From each signal, key statistical indicators were extracted, including root mean square (RMS), skewness, kurtosis and crest factor, to capture signal characteristics that were relevant to fault detection. To follow-up, we applied the Kolmogorov–Smirnov test to assess data normality and the results confirmed significant deviations from a Gaussian distribution, motivating the use of robust regression techniques for further investigations. The regression model created incorporated rotational speed and alignment conditions as predictors of acceleration and the results indicated that while the coefficient associated with misalignment suggested a possible increase in acceleration (~1.115 units), statistical testing (p = 0.5233) indicated that neither speed nor alignment had a significant influence on the measured vibration levels within the dataset. The findings suggest that under the tested conditions, misalignment does not manifest as a strong linear change in acceleration magnitude, and the study underscores the importance of robust modeling techniques and feature selection in the condition monitoring of rotating machinery.